Peer effects on health-related behaviours of adolescents in Scotland: a cross-sectional study

Sunbelt Conference, Edinburgh, 2024

Dylan Lewis, Srebrenka Letina, Babak Mahdavi Ardestani, Emily Long, Mark McCann

Social & Public Health Sciences Unit, University of Glasgow

Motivation

  • Changing health behaviours amongst young people

    • Declining alcohol and tobacco consumption

    • Growing prevalence of poor mental health

Have patterns of peer level clustering of health behaviours changed?

Do the same relationships between network structure and health behaviours observed in older datasets occur in a contemporary Scottish adolescent population?

Sample

Net4Health Peers and Levels of Stress (PaLS)
2022/23 2005/6
Cross-sectional (wave 1 of longitudinal study) Cross-sectional
4 secondary schools + pilot school 22 schools
Secondary 2 (age 12-13) and Secondary 4 (age 14-15) Secondary 4 only (age 14-15)
Online survey Pen & paper

Measures

  • Dependent variables:

    • General mental health (GHQ-12 caseness)

    • Self-esteem (Rosenberg scale, binarised)

    • Smoking (monthly or more often)

    • Drinking (weekly or more often)

    • Drugs (used this year)

  • Individual attributes:

    • Gender

    • Family affluence (FAS)

    • Parental care, parental control (Bonding Instrument BF)

  • Network ties

    • Net4Health:

      • Asked to nominate up to ten friends.

      • Two year-group networks per school = 10 networks.

    • PaLS:

      • Asked to nominate up to six friends.

      • 22 networks.

Prevalence

Example networks - Smoking

PaLS

Net4Health

Example networks - Drinking

PaLS

Net4Health

Example networks - Mental health

PaLS

Net4Health

ALAAMs

We used autologistic actor-attribute models1 (ALAAMs) to examine how the behaviour (nodal attribute) of an individual is related to their position in a social network and by the behaviours/attributes of other actors in that network.

  • ALAAMs are an extension of ERGMs.

  • Can be interpreted similarly to a logistic regression, but taking into account network dependencies.

  • Sometimes called social influence models.2

    • While we can include ‘contagion’ parameters, we can’t distinguish between influence and selections!

Plan of analysis

Missing data

  • Attributes imputed with mice R package
  • Missing DVs handled by ALAAM

ALAAMs

  • Run in R using bayesALAAM 1
  • 2 models for each network
    • A: just structural parameters
    • B: structural parameters + attributes

Meta-analysis

  • Model estimates combined in mixed-effects meta-regression using metafor R package
  • Study effect (Net4Health or PaLS) included as a moderator

Model specification

Initial results

A couple of caveats…

  • These are the results from a single imputation

  • Only structural parameters included so far (‘model A’)

  • Work in progress!

Results - Drinking

Results - Smoking

Results - Poor mental health (GHQ)

Conclusions

  • Some evidence for peer-level clustering of health-related behaviours

    • Smoking behaviours in both time-periods, drinking only in PaLS (2005/6)

  • Poor mental health was not strongly related to network structure at either time-period

  • Despite changing prevalence and societal contexts of health-related behaviours, network dependencies do not differ greatly between the time-periods

Thank you!

Questions:

d.lewis.1@research.gla.ac.uk

srebrenka.letina@glasgow.ac.uk